基于DDPG算法的无人船避障路径规划  

Obstacle avoidance path planning of unmanned surface vessels based on DDPG algorithm

作　　者：杨淞匀 王杭先 林鹏 YANG Song-yun;WANG Hang-xian;LIN Peng(School of Electronic and Information Engineering,Nanjing University of Information Science and Technology,Nanjing 210044,China)

机构地区：[1]南京信息工程大学电子与信息工程学院,南京210044

出　　处：《信息技术》2025年第3期1-7,15,共8页Information Technology

基　　金：国家自然科学基金资助项目(62201271)。

摘　　要：针对静态和动态环境下无人船路径规划问题,文中提出一种基于深度强化学习的避障路径规划方法。首先,建立水平面无人船运动模型,将路径规划描述为一个基于马尔科夫决策的强化学习过程。接着,构建基于Actor网络和Critic网络的深度确定性策略梯度算法,设计基于前视声呐测距的状态空间、运动约束下动作空间以及用于评估当前动作的复合奖励函数,并利用速度、角度、避障及动作奖励引导无人船训练神经网络。仿真结果和综合对比验证了所提强化学习路径规划方法的有效性和优越性。研究成果可为无人船在静态和动态障碍约束下的路径规划提供参考。Focusing on the path planning of unmanned surface vessels(USVs)under static and dynamic environments,an obstacle avoidance path planning based on deep reinforcement learning method is developed.Firstly,the motion model of USVs within horizontal plane is established,and the path planning is described as a reinforcement learning process based on Markov decision.Then,the Actor and Critic network based deep deterministic policy gradient algorithm is proposed,where states based on looking forward sonar ranging,actions under motion constraints and rewards for evaluating the current action are designed.Besides,the neural networks are trained by using the velocity,angle,obstacle avoidance and action reward guidance.Simulation results and comprehensive comparison verify the effectiveness and superiority of the proposed reinforcement learning path planning method.The research results can provide reference for path planning of USVs under static and dynamic environments.

关 键 词：无人船 路径规划 避障 深度强化学习 深度确定性策略梯度 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]